2 5 . , 1 * 3 $ 3 ( 5 1 2 - csef
TRANSCRIPT
&&((11775522 667788'',, ,,11 ((&&22112200,,$$ (( )),,11$$11==$$
&&((117755(( ))2255 667788'',,((66 ,,11 ((&&22112200,,&&66 $$11'' )),,11$$11&&((
::2255..,,11** 33$$33((55 1122�� ����
Risk Sharing and the Tax System
Charles Grant and Mario Padula
August 2001
DIPARTIMENTO DI SCIENZE ECONOMICHE - UNIVERSITÀ DEGLI STUDI DI SALERNO Via Ponte Don Melillo - 84084 FISCIANO (SA)
Tel. 089-96 3167/3168 - Fax 089-96 3169 – e-mail: [email protected]
::2255..,,11** 33$$33((55 1122�� ����
Risk Sharing and the Tax System
Charles Grant* and Mario Padula**
Abstract Several papers have documented that US consumers can not fully insure themselves against all their idiosyncratic risks, but little is understood about which mechanisms provide insurance. We investigate whether, as some suggest, progressive taxes provide additional insurance. The methodology distinguishes insurance from redistribution, and can by applied to testing any potential insurance mechanism. Using repeated cross-sections from the US consumer expenditure survey (CEX), we relate changes in consumption inequality to several measures of tax progressivity. Identification exploits the variation in taxes both across states and over time. Our results suggest, under weak assumptions, that progressive taxes do not induce insurance, while stronger assumptions quantify this effect. Acknowledgements: We would like to thank Orazio Attanasio, Christos Koulovatianos, Tullio Jappelli, Luigi Pistaferri, Oved Yosha and several seminar participants for their helpful comments and advice. All remaining errors are, of course, our own. The TMR Network Program on Saving and Pensions, the Italian National Research Council (CNR) and the Ministry of University and Scientific Research (MURST) provided financial support. Address for correspondence: [email protected]
* University College London and CSEF, University of Salerno. E-mail address:[email protected]
** CSEF, University of Salerno and University College London. E-mail address: [email protected]
Contents
1 Introduction 6
2 Taxes and Transfers 8
3 Empirical Framework 10
3.1 Quadratic Utility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.2 Alternative Utility Speci�cations . . . . . . . . . . . . . . . . . . . . 13
3.3 The Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4 Data 18
4.1 Measuring Tax Progressivity . . . . . . . . . . . . . . . . . . . . . . . 22
5 Results 23
6 Conclusions 27
5
1 Introduction
Over the last twenty years or more, there has been considerable interest, particularly
among consumption economists, about the extent to which consumers can smooth
out income shocks. While the earlier literature considered consumers smoothing over
time, the much of the more recent literature has looked at smoothing across states
of nature, asking whether households can fully insure their idiosyncratic risks. An
implication of the full insurance hypothesis is that neither temporary nor permanent
shocks to income will have any impact on consumption behaviour. While this hy-
pothesis has consistently been rejected by the literature (for US consumers, see for
instance Mace 1991, Cochrane 1991, or Attanasio and Davis 1996), it seems unlikely
that agents are completely autarkic: more likely, they have some, but partial, ability
to smooth shocks. A natural corollary is to ask what mechanisms provide insurance?
This has important welfare implications, since providing additional insurance, other
things remaining equal, will make consumers better o�. Thus we investigate what
changes the total amount of insurance available to consumers, explicitly recogniz-
ing that any potential mechanism might merely crowd out, or substitute for, other
risk-sharing mechanisms.
One particularly popular candidate instrument that might provide insurance is
the tax and bene�t system. This is explicitly recognized in some tax regimes: for
instance, in Britain part of the tax system is labeled national insurance. Moreover,
some economists, such as Varian (1980), have argued that this is a more important
motivation for progressive taxes than redistribution. Several papers have studied this
problem, including Attanasio and Rios-Rull (2000), and Kruegar and Perri (1999),
by simulating over some plausible parameter estimates. Never-the-less, whether the
tax system really does provide insurance, or merely crowds out, or substitutes for,
other forms of insurance is an open issue.
In an interesting and much discussed paper, Asdrubali, S�rensen and Yosha (1996)
investigate how state level shocks are insured, or smoothed, across US states. Their
6
exercise decomposes insurance into its various constituent parts, by regressing changes
in consumption against changes in the components of income. They �nd that federal
taxes smooth 13% of income variation, but their results suggest that taxes are only
half as important as credit markets, and a third as important as capital markets (e.g.
holding a balanced portfolio of assets), in smoothing state level consumption. Instead,
this paper looks at individual level shocks, and investigates the level of insurance
between households within the state. Our approach has a number of advantages: by
using individual data we can separate insurance from redistribution, something they
concede they can not distinguish. We also want to know the marginal e�ect. We have
in mind the following policy experiment: suppose the tax system was changed so that
it became more redistributive; will the level of insurance change, or will consumers
merely substitute from alternative insurance mechanisms. In the decomposition of
Asdrubali et. al. (1996) this e�ect is not captured.
The methodology of this paper builds and develops the earlier work of Deaton
and Paxson (1994). We exploit di�erences across US states, as well as time, to
identify what e�ect, if any, the tax system has on the total amount of insurance
available to consumers: identi�cation follows from the fact that di�erent US states
have di�erent tax regimes. We use household level consumption data available from
the CEX for the years 1982-1997. The aim is not to reject full insurance, but rather
to test whether, and how much, a progressive tax system provides insurance. For
this, as will be explained, it is not necessary to explicitly state what is observed in
the full insurance case, unlike the paper by Deaton and Paxson (1994). While quite
general assumptions can be used to identify whether taxes provide insurance, tighter
assumptions are needed to quantify how much is provided. We will �nd that the tax
system does not help agents to insure themselves against income shocks.
The paper is organised so that �rst it will discuss how a progressive tax system
might provide insurance. Then it discusses the literature on how to test for (complete)
insurance, before explaining how the literature can be used to test for the marginal,
7
or extra, amount of insurance that some policy instrument, such as the tax system,
might provide. We then brie y discuss our data, and how the tax system works in
the US, before we present the results of our regressions. Finally after discussing our
results, we brie y conclude.
2 Taxes and Transfers
It has been traditional in the public economics literature to view taxation as a method
of reducing cross-sectional inequality. The idea is that redistributing income from
high to low income households may be a 'good thing'. Obviously, unless agents
were altruistic, such a plan will be supported by low income people and opposed
by high income people. If, on the other hand, agents were altruistic, there would
seem to be no motivation for a public scheme of re-distribution, since agents would
be privately motivated to redistribute their income. However, this argument ignores
another (possible) important motivation for a redistributive tax.1 If future income is
uncertain, then, in the absence of alternative insurance mechanisms, a redistributive
tax system can provide insurance against this uncertainty. Varian (1980), for instance,
argued that this ought to motivate extremely high marginal tax rates.
Suppose individual i's current income yit has a permanent ypit and a temporary
component2:
ypit = ypit�1 + fit
yit = ypit + "it
(1)
This periods income is subject to permanent fit and temporary uit shocks, which are
assumed to be mean zero processes. Thus income can be written:
yit = fi0 +tX
s=1
fis + "it (2)
1Here a poll tax is deemed non-redistributive, and any tax that increases as income increases is
deemed to be redistributive, or progressive.2This analysis is similar to that presented in Deaton et. al. (2000)
8
Taxes are typically a function of current income: suppose taxes are levied at some at
rate level � . In which case expected per capita tax revenue (where the expectation is
over the i individuals) is:
�Ei
"fi0 +
tXs=1
fis + "it
#= �Eifi0 = � �fi0
Suppose these taxes are re-distributed back to all agents as a lump sum transfer.3
After tax and transfer income, ~yit, would be:
~yit = fi0 � ��fi0 � �f0
�+ (1� �)
"tX
s=1
fis + "it
#(3)
and the change in income can be decomposed into an insurance component and a
re-distribution component.
yit � ~yit = ��fi0 � �f0
�| {z } + �
hXfis + "it
i| {z }
redistribution insurance
(4)
Re-distribution is on the initial distribution of income, while insurance acts on any
income shocks. We concentrate on this second feature of the tax system. Whether,
in practise, the tax system provides insurance depends on what mechanisms the
agent has to smooth consumption in the presence of income shocks. If no other
insurance mechanisms were available, then clearly a progressive tax system would
provide insurance. On the other hand, the agent may already have alternative risk-
sharing arrangements available to him, in which case the tax system may instead
interrupt the agents private incentives to participate in these arrangements. In this
case, a progressive tax system may provide no additional insurance, but may instead
merely crowd out, or substitute for, existing arrangements, a point highlighted by
Attanasio and Rios-Rull (2000).
3It would be simple to adjust this equation if some �xed proportion of �fi0 is spent by the
government on public goods, rather than being redistributed. This adjustment has no e�ect on the
analysis.
9
We propose using the di�erent tax systems in the di�erent states of the US to
examine how the level of insurance changes as the degree of re-distribution changes.
However, rather than attempting to simulate over the possible values of the parame-
ters in a fully structural model, an approach taken in Attanasio and Rios-Rull (2000)
or Deaton, Gourinchas and Paxson (2000) among others, we will instead estimate a
reduced form, and see what insights can be gained from that approach. While the
structural approach can give valuable insights, we do not believe that it can success-
fully answer the question that we are interested in: it would require knowledge of
preferences, the income process, and all insurance mechanisms available to the agent.
This is not necessary in the reduced form approach that we take in this paper. While
it can not describe the mechanism that provides insurance, it can say whether insur-
ance is provided. This is often what a policy maker wants to know. The aim is to
compare di�erent tax systems to see how the overall level of insurance changes across
regimes.
3 Empirical Framework
A number of approaches have been used to test for full insurance. The earliest tests
are based on exclusion restrictions. Suppose each individual i's income yit was subject
to aggregate �t and idiosyncratic uit shocks each period t so that
yit = yit�1 + �t + uit
By construction, the idiosyncratic component uit sums to zero over the i individu-
als. Full insurance implies that this term does not a�ect changes in consumption,
hence a valid test for full insurance is to regress changes in consumption �cit on the
idiosyncratic component of the income shock. Mace (1991), Cochrane (1991), and
Attanasio and Davis (1996) all apply this test to US households, and decisively reject
this implication of the full insurance hypothesis. A second approach, due to Jappelli
and Pistaferri (1999), instead tests the ordering of agents. Under full insurance, if
10
agents, in any time period, are ordered by their level of consumption, then this or-
dering (after controlling for taste-shifters) will not change over time. They reject full
insurance, and argue that neither measurement error, nor taste-shifters, could explain
this rejection.
A third implication was investigated by Deaton and Paxson (1994). For the mo-
ment ignore taste-shifters, then, with quadratic utility (and assuming the interest
rate r equals the discount rate Æ) consumption follows a martingale:
cit = cit�1 + "it (5)
in which case the cross-sectional variance of consumption for a �xed membership
group j evolves according to the relationship:
varj (cit)� varj (cit�1) = varj ("it) + 2covj (cit�1; "it)
Deaton and Paxson (1994) showed that if lagged aggregate consumption is in each
agent's information set then covj (cit�1; "it) = 0 (at least on average over a large
enough number of time periods). If income follows the stylized process:
ypit = ypit�1 + fit
yit = ypit + vit(6)
then, for quadratic utility, the change in consumption is due to the permanent shock,
and the annuity value of the transitory shock, i.e.:
"it = fit +r
1 + rvit
Blundell and Preston (1998) used this to show, in their proposition 3, that:
�varj (cit) = varj (fit) +�
r
1 + r
�2varj (vit) (7)
If the permanent and temporary shocks are fully insured, then neither should a�ect
consumption, which would imply:
�varj (cit) = 0
11
This implication of full insurance was rejected by Deaton and Paxson (1994). An
advantage of their approach, compared to the other two, lies in the weaker data
requirements. In order to test for full insurance, researchers need only observe the
cross-sectional moments of the distribution of consumption. In contrast, the �rst
approach requires controlling for idiosyncratic versus aggregate income shocks, which
unlike in the Deaton and Paxson case, may not be feasible with a time-series of cross-
sections, while the second approach requires the availability of panel data, and for all
taste-shifters to be fully parametrized. For these reasons we build from the approach
of Deaton and Paxson (1994).
While full insurance has been consistently rejected by all of the approaches dis-
cussed above, the tests are not fully constructive, in the sense that they do not help
us to understand either how much insurance is available to agents, nor what mecha-
nisms might change the overall level of insurance. The next two subsections discuss
incomplete insurance under quadratic and alternative utility speci�cations.
3.1 Quadratic Utility
Under quadraric utility, with r = Æ, the permanent income hypothesis implies:
cit = cit�1 + fit +r
1 + rvit
If insurance is incomplete, so that the agent can insure the proportions � of his
permanent shock, and of the temporary shock then:
cit = cit�1 + (1� �) fit + (1� )�
r1+r
�vit
) �varj (cit) = (1� �)2 varj (fit) + (1� )2�
r1+r
�2varj (vit)
where the variance is taken for a �xed membership group j. In this framework, � and
are determined by the characteristics of the economy in which the agent resides.
They re ect all the mechanisms available to the agent that can insure idiosyncratic
risk. Two cases are of particular interest: if � = = 1 then this implies full insurance,
12
while � = = 0 re ects autarky, where none of the agents idiosyncratic risk can by
insured. Further, if r is suÆciently small, or if the temporary shock can be fully
insured ( = 1), then4:
�varj (cit) = (1� �)2 varj (fit) (8)
From this last equation two things are immediately apparent. First, unless � = 1,
the full insurance case, the variance (or standard deviation) of consumption will be
increasing over time. Secondly, as � increases (the amount of insurance increases) the
growth in the variance of consumption falls.
3.2 Alternative Utility Speci�cations
For more complicated utility functions the simple relationship in equation 8 no longer
holds. One of the simplest extensions is discussed by Attanasio and Jappelli (2001).
Maintaining quadratic utility, but relaxing r = Æ, results in the Euler equation:
cit =
1 + Æ
1 + r
!cit�1 + "it (9)
where, under the assumptions made above on the income process, consumption inno-
vations "it are equal to a linear transformation of income shocks, i.e.:
"it = fit +r
1 + rvit
And under full insurance varj (cit) may be growing or declining depending on whether
r > Æ or r < Æ. With incomplete insurance and under the assumptions made above
on income shocks the evolution of the variances for group j takes the form:
varj (cit) =�1+Æ1+r
�2varj (cit�1) + (1� �)2 varj (fit) (10)
where the assumption that idiosyncratic shocks are all smoothed away is maintained.
4This assumption may not be too unreasonable. Attanasio and Davis (1996) o�er evidence that
temporary shocks can be smoothed by US consumers, but permanent shocks can not.
13
Attanasio and Jappelli (2001) also discuss the case of isoelastic utility functions
with relative risk aversion parameter �. In this case, consumption innovations are not
a simple function of income shocks, as in the quadratic utility case. If the distribution
of consumption growth was log-normal, then the Euler equation takes the form:
ln cit = ln cit�1 + � (r � Æ) +1
2�var (� ln cit) + "it
If the �xed membership group j is suÆciently homogeneous, so that all members faced
the same uncertainty about consumption growth, then the variance of consumptions
evolves according:
varj (ln cit) = varj (ln cit�1) + (1� �)2 varj ("it) (11)
Notice that the last term of the RHS of equation 11 is the variance for group j of
the consumption innovation and that full insurance would now imply that there is no
change in the variance of log-consumption over time.
Blundell and Preston (1998) discuss the cases of constant absolute risk aversion
(CARA) and constant relative risk aversion (CRRA) preferences. For the �rst they
use a relationship derived by Caballero (1990) who �nds that when income is gen-
erated by a random walk and income shocks, "it, are log-normally distributed the
growth of consumption is:
� ln cit = �it + �it + "it
where �it is the slope of the consumption path and �it is a term that accounts for
revisions to variance forecasts. Again, if the �xed membership group j is suÆciently
homogeneous, so that �it and �it are constant within each group, then under incom-
plete insurance, the variance of the growth rate of consumption evolves according:
varj (� ln cit) = (1� �)2 varj ("it) (12)
14
Blundell and Preston (1998) also showed that a relationship of the same form as
equation 12 holds when preferences are described by a CRRA functions and the
income process is given by:
ln yit = ln ypit + vit
ln ypit = ln ypit�1 + fit
Namely, it holds that:
varj (� ln cit) = (1� �)2 varj (fit) + (1� )2�
r
1 + r
�2varj (uit) (13)
If, as before, we assume that either the interest rate is suÆciently small, or that
� 1, equation 13 boils down to equation 12.
Lastly, it is worth considering completely arbitrary preferences, that nevertheless
maintain that the real interest rate and the discount rate are equal. Changes in
marginal utility must obey the relationship:
� (cit) = � (cit�1) + �it
where � (�) is the marginal utility of consumption and �it is its innovation at time t.
Then the variance of the marginal utility consumption for group j under incomplete
insurance evolves according to:
varj (�it) = varj (�it�1) + (1� �)2 varj (�it) (14)
which implies that if � 6= 1 the variance of the marginal utility of consumption is
increasing over time. There are two problems in bringing 14 to the data. First, the
variance of the marginal utility can be computed only if one makes speci�c assump-
tions on preferences. Second, the variance of the innovation in the marginal utility of
consumption is not observable.
In the following, we show how to tackle the �rst problem, while the next subsec-
tion shows how our identi�cation procedure takles the second problem. The Euler
15
equation can be used to characterize the changes in the variance the marginal utility
of consumption. A �rst order Taylor expansion around cit�1 gives:
�it = �it�1 + �0it�1 (cit � cit�1) + o (cit � cit�1) (15)
Hence the variance of (15) is:
varj (�it) = varj (�it�1) + Ej��0it�1
�2varj (�cit)
+varj��0it�1
�Ej (�cit)
2 + varj��0it�1
�varj (�cit)
(16)
where Ej (�) is the expected value for �xed membership group j. From equation
(16) one can see that if the variance of the change in consumption increases then the
change in the variance of the marginal utility of consumption also increases. This
implies that those factors that cause the variance of the change in consumption to
decrease cause the change of the marginal utility of consumption to decrease too, i.e.
the overall amount of insurance available to individuals to increase. This allows us to
identify what provides insurance at the margin, even without making the assumptions
about the functional form of the utility function. In other words, there is a monotone
mapping between the changes in the variance of marginal utility of consumption and
the variance of changes.5
3.3 The Regression
The discussion above highlights that the sign of the e�ect of a policy instrument,
as captured by �, can be always identi�ed, while the coeÆcient is interpretable only
under speci�c assumptions about the utility function. In the discussion on quadratic
utility above, we considered �varj (cit), but if, in equation 5, we �rst di�erence before
taking variances, we obtain:
varj (�cit) = varj ("it) (17)
5One could obtain an entirely similar derivation using logs instead of levels.
16
and the implications are the same. However, in this approach, no assumption about
the covj (cit�1; "it) is necessary, although the disadvantage is that each household
must be observed at least twice. The discussion on alternative utility speci�cations,
culminating in equation 16, also suggests using the variance of changes rather than
changes in variances. The third reason for doing this is it allows us to easily measure
the insurance e�ect of our policy instrument. This last follows since equation 8
becomes:
varj (�cit) = (1� �)2 varj (fit) (18)
which implies that if instead the standard deviations were taken then:
s:d:j (�cit) = (1� �) s:d:j (fit) (19)
Recall that � is the amount of insurance available to households within each regime.
This amount may depend on some policy instrument zj which varies exogenously
across regimes. In which case, estimation can recover � as a function of zj. Variation
between regimes allows this function to be recovered. Writing:
��zj�= �0 + �1z
j + �j (20)
where the error term �j captures any di�erences in the level of insurance that are not
modeled. Full insurance implies that both �0 = 1 and �1 = 0; incomplete insurance
implies �0 < 0; while zj provides additional insurance if �1 > 0. If instead zj merely
crowded out, or substituted, for other insurance mechanisms, then this would imply
�1 = 0, while a negative coeÆcient would imply that there was more than complete
crowding out (that is the overall level of insurance is reduced). Substituting into
equation 19 results in:
s:d:j (�cit) =�1� �0 � �1z
j � �j�s:d:j (fit) (21)
The regression that is run takes the form:
s:d:j (�cit) = �0 + �1zj + error (22)
17
and identifying terms implies:
�0 = (1� �0) s:d: (fit)
�1 = ��1s:d: (fit)
error = ��js:d: (fit)
Hence if the variance of the permanent shock were known and constant across all
groups j, the level of insurance, and how it changes with the policy instrument zj,
can be recovered.
For departures from the simple cases, the coeÆcient is no longer so easily identi�ed
but, the sign can always be interpreted. This is illustrated in �gure 1. The top panel
shows the simple quadratic case where r = Æ. Full insurance implies that varj (cit) is
time-invariant, while full risk-sharing is rejected if the variance is growing over time.
This is longer true for more general preferences: instead full risk-sharing may imply
that varj (cit) is either increasing or decreasing, depending on the parameters of the
model. The diagram shows the variance decreasing under full insurance. However,
our aim is to compare the di�erent policy regimes, indexed by zj, hence we compare
the varj (cit) across the di�erent regimes: in those regimes where this variance is
growing more slowly, there must be more risk-sharing or insurance. This remains
true even if the full insurance case can not be fully speci�ed.
4 Data
This paper uses the Consumer Expenditure Survey (CEX): a survey of US households
that has operated on a continuous basis since 1980. The data not only has detailed
information on consumer expenditure, income and taxes, it also records information
about the state of residence of the household (although, due to con�dentiality re-
quirements, this information has been suppressed for some observations). Comparing
these states will enable us to test whether the tax (and transfer) system can provide
insurance as di�erent states have di�erent tax policies. This paper uses a sample
18
Figure 1: The evolution of varj (cit).
age=time
varj (cit)
���������������������������
full insurance
rejectfull insurance
(a) Quadratic preferences where r = Æ.
age=time
varj (cit)
hhhhhhhhhhhhhhhhhhhhhhhhhhh
���������������������������
full insurance
z1
z2
(b) A more general case, comparing regimes z1 and z2.19
of around 100,000 households taken from the years 1982 to 1997 for which we have
full state information. The data was also restricted to those households headed by
individuals who were between 25 and 55 in the �rst year in which they are sampled.6
In the US taxes are raised at a variety of levels; those entitled to levy taxes include
the federal and state legislators, county administrations, and school boards. Taxes in-
clude income taxes, sales taxes, property taxes and duty. This paper will concentrate
on income tax, which is raised both federally and by states. While the consumer ex-
penditure survey includes questions on the amount of taxes that the household pays,
we do not believe that the answers that households give are particularly accurate or
reliable. Nevertheless, some results are included using these variables. However, we
have also constructed a tax liability �gure using the TAXSIM programme developed
by Freenberg (see Freenberg and Coutts, 1993 for details) and provided by the NBER
and including state taxes since 1982. Using a variety of household variables, includ-
ing husband's and wife's salary income, taxes and other costs on property, interest,
dividends etc., and details about the household's characteristics (such as number of
children) as well as the state of residence, the programme constructs both the state
and the federal tax bracket, tax liability, and marginal tax rate for each household
in the sample. The calculation speci�cally allows for the fact that a variety of al-
lowances are allowed, and also includes an assessment of households entitlement to
the Earned Income Tax Credit7 (which results in some households net tax liability
being negative). Tables 1-5 summarize some of the main features of the data.
The current federal marginal tax rate (see table 1) varies from 15% for those whose
income is less than $26,250 (for single people, for married couples the threshold is
$43,850) up to 39.6% for income over $288,350. Table 2 summarizes the proportion
of people in each tax bracket in our sample. It also highlights that the brackets
themselves have varied over the years. For the period 1982-1986 a large number
6This means if we are taking the k-th di�erence they were aged 30+k to 50+k in the last year in
which they were sampled.7See Scholz (1996) for an explanation of the EITC.
20
of tax brackets were applicable: too many to give anything other than fairly broad
summary statistics. However, the table demonstrates that around 15% (depending
on the year) did not pay any tax, while the median household was in the 23% tax
bracket. In these years the highest tax bracket for federal taxes was set at 50%. In
1987 the number of brackets was greatly reduced, and from this year every tax bracket
is recorded in table 2. There was a further reduction in the following year. Between
1987 and 1996 the proportion of households who did not pay tax gradually increased
to 19%. Other features are the introduction of a 31% tax bracket, for the top 5%
of earners in 1991, the introduction of a 36% bracket in 1993, of a 39.6% bracket in
1996, and the abolition of the 0% bracket in 1997.
States taxes can di�er quite widely among the di�erent US states. Table 3 shows
the current tax rates applicable in di�erent US states. From this we can see 8 states,
including Texas and Florida, do not to levy any income tax on their residents: state
revenue in these states comes mostly from sales taxes. In addition New Hampshire
and Tennessee only charge state income tax against dividend and interest income.
The other states have a variety of income tax bands and exemptions (or tax credits)
that are applicable. Although some states have a at rate tax, in most states, the
marginal tax rate increases with income, and there are a variety of tax allowances to
which households are entitled. Table 3 provides a summary description of the state
tax systems in the U.S.
The paper also exploits the transfers that typically poorer agents receive. Such
transfers include social security and railroad retirement income, supplementary se-
curity income, unemployment compensation, worker's compensation and veterans
payments, public assistance or welfare, pension income, and the value of food stamps
received: the CEX includes questions on all these transfers. Table 4 shows that the
average amount of transfer, over the whole sampled population, is $871, but that only
18.6% of households receive a transfer. Conditional on receiving at least something,
households receive an average of $4,680. This is not a substantial amount, when the
21
average household salary income in the survey is $34,574.
4.1 Measuring Tax Progressivity
In order to assess if tax systems are a�ecting the level of insurance available to
households we need some measure of the progessivity8 of the tax system in each
state. If the marginal tax rate were the same for all households, then this could be
used in the regressions. As it increases we should expect the cross-sectional variance
of consumption to increase. The intuition is that higher marginal tax rates cause the
after tax income distribution to be more concentrated. However, from the previous
discussion, the marginal tax rate that consumers face is an increasing function of
income, and furthermore, households di�er in the allowances and exemptions they
can claim. Under such circumstances, no completely satisfactory measure of tax
progressivity exists. We choose to use two measures that capture progressivity. The
�rst is to take the average of the households marginal tax rate, or tax bracket, within
each year-state-cohort j. From table 5 the average federal bracket is 20.2%, and the
average of the reported marginal tax rates (which accounts for various allowances) is
19.2%. The state rates vary from zero in Texas and Florida, which charge no taxes,
to an average marginal tax rate of 7.4% in New York.
One problem with using this measure is that it ignores any heterogeneity in tax
rates across households. For instance, a given mean marginal tax rate of 20% could
be due to all households paying a marginal (and average) tax rate of 20%; from only
the top half of the income distribution paying, but paying 40%; or from only the
bottom half paying 40%. These situations di�er strongly in the degree to which taxes
are re-distributive. A related problem, is that the mean marginal tax rate is not
invariant to the overall level of taxes that are taken. Our interest is in the degree to
which taxes re-distributes income, hence we wish to construct a measure that is not
8Recall that we de�ne a tax to be progressive if the tax liability increases as income increases,
thus a at rate income tax is deemed progressive.
22
a�ected by the overall tax level. This motivates a second measure of how much the
tax system redistributes income, constructed as:
� = 1�
vuutvarj (~yit)
varj (yit)(23)
that is, as one minus the square-root of the ratio of the variance of income after taxes
compared to before taxes for each group j. If all households faced the same marginal
tax rate, and there were no allowances, then this constructed � would exactly equal
the marginal tax rate. Moreover, a larger � implies more re-distribution. Table 5
displays the constructed values for � for the whole of the US and for the 6 largest US
states. It shows that this measure averages to 32.0% over the US, but that states can
di�er from 27.6% in Florida (where there is no income tax), to 36.8% in New York,
traditionally viewed as one of the more progressive states. This measured � can be
regressed on how much the variance of consumption changes over time. A negative
coeÆcient implies that the variance is growing more slowly, which is interpreted as
meaning more insurance is being provided by the tax system. That is, a negative
coeÆcient means that as taxes become more re-distributive, agents are better able
to smooth against uncertain income shocks. In contrast, a positive coeÆcient implies
that taxes more than crowd out alternative insurance mechanisms, and the overall
level of insurance from the tax system is reduced.
5 Results
This section discusses the results. Recall that section 3 motivated regressing the
change in the standard deviation of consumption against a measure of tax progres-
sivity, � . The regression takes the form:
s:d:j (�cit) = �0 + �1�j + "j
where j represents the state-time-cohort combination, and we di�erence it over nine
months (the largest possible period given our data). The number of groups that could
23
reasonably be de�ned (ensuring that the cell size was at least 75) was around 180,
but depended on the regression.9 Obviously, this process meant that several smaller
states were never included in the analysis, since fewer observations came from these
states. However, there were suÆcient observations in most of the larger states.
Tables 6-9 display the results. In order to minimize the e�ect of sample compo-
sition, we take for each state only those households who are between 25-40 years old
and those who are between 41-55 at the beginning of the sample. In column 1 of each
table we report our baseline speci�cation, where we only control for seasonality in
consumption. Moreover, since the variance of the growth rate of households is likely
to be a�ected by heterogeneity within the group, which is at large extent predictable,
the regression ran in column 2 add a number of controls to the baseline regression.
That is, instead of using s:d:j (�cit), column 2 uses s:d:j [�cit � E (�citjXit)]. The
set of control variables X includes age, time, education, sex, race, marital status, and
changes in family size.
A second important issue is that we wish to observe the standard deviation of
consumption, and for the measure of compression for the whole population in each
group, but we only observe a small sample of the population. Thus errors are intro-
duced into both the left-hand side and into the right-hand side of each regression.
That is, we have:
s:d:j (�cit) + "jt = �0 + �1�� jt + &jt
�+ uj
t (24)
One would expect past values of � jt to be correlated its current value in its state, but
measurement error induced by the small sampling sizes to be uncorrelated with the
past, hence instrumenting with lagged values of � jt will remove any downward bias. A
further problem is that "jt and &jt may be correlated, hence for some of the regressions
� jt twice lagged will be the instrument. Columns 3,4,5 of each table host IV estimates,
where the instruments are � (or the marginal tax rate) lagged once, lagged once or
9Experimenting with di�erent cell sizes did not qualitatively change the results. We also experi-
mented with di�erent ages, di�erent cohorts, and di�erent years, all without substantive e�ect.
24
twice, or lagged twice. The signi�cance of the results is starred once for signi�cant
at the 10% level, twice for signi�cant at the 5% level and three times for signi�cant
at the 1% level.
Table 6 shows that in the basic regression, and in the regression including a set of
control variables, that the coeÆcient on the mean marginal tax rate, and on the tax
bracket, is signi�cant at the 1% level. However, perhaps surprisingly, the coeÆcient is
positive. This suggests that other important insurance mechanisms are being crowded
out by the introduction of a more re-distributive tax system. When the instrumented
results are included, the mean marginal tax rate remains signi�cant, at the 5% level,
although this is not true for the mean tax bracket. When the change in the variance of
log-consumption10 was considered in table 7 the coeÆcient on the constant becomes
highly signi�cant (at the 1% level), and the marginal tax rate, and the tax bracket
both remain signi�cant. This remains true when � is instrumented by its lag, and
even in the last column, where it is lagged twice. The results in this last column are
signi�cant at the 1% level.
Tables 8 and 9 consider the constructed value of re-distribute that was created
using before and after tax income. The �rst results refer to levels, and show, except in
the less reliable case where we have used the reported tax liabilities in the CEX, that
� is positive and signi�cant at the 5% level. However, when � is instrumented by past
values of the variable, the results are more ambiguous, and only remain signi�cant for
the middle panel that ignores transfer income. When log-consumption is considered
the results are signi�cant at the 1% level, with and without the controls and for both
� and the constant. The signi�cance of the results remain with the IV-estimate. The
last column, our most preferred regression, is also signi�cant.
The fact that the constant was signi�cant and positive in the regressions, at least
for the log-consumption regressions, con�rms the results in Deaton and Paxson (1994).
This rejects full insurance, although it is open to the criticism made in Attanasio and
10This is a better measure of risk-sharing when preferences are not quadratic.
25
Jappelli (2001). The results for the coeÆcient for the measure of tax-progressivity
have a number of interpretations. At the most basic, the coeÆcient on � can be
treated as an exclusion restriction: under full insurance the growth in the variance of
consumption, regardless of the utility function, should be uncorrelated with anything
except changes in tastes. Hence the fact that � enters signi�cantly rejects the full
insurance hypothesis. A second interpretation says that while the coeÆcient is not
interpretable, the fact that the coeÆcient is positive, and signi�cant, means that
making the tax system more progressive is reducing the amount of insurance that
the agents have. There must be other mechanisms that operate privately, that are
being disrupted by the imposition of the tax system: agents incentives to participate
in these private insurance mechanisms is reduced since they have access to a public
mechanism. However, how these private mechanisms operate is not calculated.
The �nal interpretation allows us to quantify how much dis-insurance the tax
system provides. The theory section highlighted when this case arises and the terms
can be identi�ed:
�0 = (1� �0) s:d: (fit)
�1 = ��1s:d: (fit)
error = ��js:d: (fit)
For identi�cation it is necessary to know, or estimate, the standard deviation of the
permanent shock. For log-income, this has been done by MaCurdy (1982) among
others.11 His paper estimated the variance of the permanent shock to be 0.34: the
implications for the estimated dis-insurance that is thus generated by a more re-
distributive tax system is tabulated in table 10. From the table, both the basic, and
the control regression suggests, for either the marginal tax rate or the tax bracket,
that households can insure roughly one third of their permanent income shock. Each
one percent increase in the marginal tax rate reduces the level of insurance by 0.7%.
11We are not aware of any papers that have estimated this parameter when income has been
measured in levels, and hence concentrate on the log-level case.
26
Instrumenting both increases the estimated basic level of insurance and the estimated
reduction in insurance induced by the tax system. Those regressions that use the
constructed � proposed in equation 23 show similar results for the basic and the
control regression. When instruments are used the results that used the estimated
taxes from the TAXSIM routine described above show that households can insure
roughly 45% of their permanent shock, but that each 1% increase in the measure of
how re-distributive the tax system is reduces the amount of insurance by 1%. To
put this in perspective, these results imply that if we examine table 5, if a household
was moved from California to Florida, then the total amount of the permanent shock
that the agent could insure would increase from around 9% to around 18%, that is it
would roughly double.
6 Conclusions
In this paper we tried to relate the amount of insurance available to households to the
system of taxes and transfers. A large strand of literature has documented the absence
of full insurance, at least for the U.S. economy. However, it is likely that households
can smooth some of their idiosyncratic shocks. Several mechanisms might provide
insurance: some of them, such as �nancial markets, are related to formal economic
interactions; others, such as family networks, relate to informal interactions. The tax
system is the mechanism this paper explores.
Taxes and transfers re-distribute income across households, but also might smooth
some of the income shocks households incur. This in turn a�ects how consumption
changes over time. If the changes in consumption respond to idiosyncratic shocks,
we should expect this response to be weaker as more insurance becomes available to
households. Moreover, if the cross-sectional variance of consumption trends up, we
should expect this trend to be atter, the more households are able to insure their
consumption against idiosyncratic income shocks. Around these two intuitions, the
27
empirical results of the paper are organized.
We test if changes in consumption respond to the degree of compression of the
income distribution induced by taxes and transfers. We measure the degree of dis-
tribution induced by taxes and transfers with (a) the marginal tax rate, and the
tax bracket; (b) a constructed measure of how much taxes compress the distribution
of current income. Our regressions show that increasing the degree to which taxes
re-distribute income is negatively correlated with the increase in the variance of con-
sumption growth. The paper suggests three interpretations: (i) this can be thought
of as an exclusion restriction that con�rms the rejection of full insurance found by
previous authors; (ii) the fact that the coeÆcient is positive suggests that increasing
the degree to which taxes re-distribute income reduces the total amount of insur-
ance available to households; and (iii) under stronger assumptions, the coeÆcients
themselves can be interpreted, and the results suggest that moving from a highly re-
distributive state, such as California, to one that re-distributes less, such as Florida,
can double the amount of insurance that is available to households.
The overall, and tentative conclusion of this paper, is that the tax system seems to
be rather a poor mechanism for providing insurance against the idiosyncratic income
shocks that households su�er. Instead, the tax system crowds out other mechanisms
which we do not attempt to describe, although the methodology described in this
paper can be extended to any potential insurance mechanism: we merely need to
observe suÆcient variation in our sample. This results suggests that any defense of
a re-distributive tax system that appeals to the self-interest of the agents involved
must argue on the grounds that ex ante inequality is itself a 'bad thing' since the
results do not support the suggestion made by Varian (1980) that progressive taxes
can be motivated by agents insurance incentives, at least for the ranges of taxes that
are observed in the US.
28
References
[1] Asdrubali, S�rensen, B., and Yosha, O., (1996), "Channels of Interstate Riskshar-
ing: United States 1963-1990", Quarterly Journal of Economics, 104, pp. 1081-
1110.
[2] Attanasio, O., and Davis, S., 1996, "Relative wage movements and the distribu-
tion of consumption", Journal of Political Economy, 104, pp. 1227-1262.
[3] Attanasio, O., and Jappelli, T., 2001, "Intertemporal choice and the cross-
sectional variance of marginal utility", The Review of Economics and Statistics,
83(1), pp. 13-27.
[4] Attanasio, O., and Rios-Rull, J., 2000, "Consumption smoothing in island
economies: can public insurance reduce welfare?" European Economic Review,
44(7), pp. 1225-1258
[5] Cochrane, J., 1991, "A simple test of consumption insurance", Journal of Polit-
ical Economy, 99, pp. 957-976
[6] Deaton, A., 1991, "Saving and liquidity constraints", Econometrica, 59, pp. 1221-
1248
[7] Deaton, A., 1993, "On risk, insurance and intra-village smoothing", preliminary
draft.
[8] Deaton, A., and Paxson, C., 1994, "Intertemporal choice and inequality", Journal
of Political Economy, 102, pp. 437-467
[9] Deaton, A., Gourinchas, P. O., and Paxson, C., 2000, "Social security and in-
equality over the lifecycle", NBER working paper no. 7570.
[10] Dercon, S. and Krishnan, P., 2000, "In sickness and in health: risk sharing within
households in rural Ethiopia", Journal of Political Economy, 108, pp. 688-727.
29
[11] Freenberg D., and Coutts E., 1993, "An Introduction to the TAXSIM Model",
Journal of Policy Analysis and Management, 12(1), pp ???
[12] Hall, R. E., 1978, "Stochastic implications of the life-cycle permanent income
hypothesis: theory and evidence", Journal of Political Economy, 86, pp. 971-
988.
[13] Jappelli, T. and Pistaferri, L., 1999, "Intertemporal choice and consumption
mobility", CSEF working paper no. 23
[14] Kreugar and Perri, 1999, mimeo
[15] Mace, B., 1991, "Consumption volatility: borrowing constraints or full insur-
ance", Journal of Political Economy, 99, pp928-956
[16] MaCurdy, T. E., 1982, "The use of time series processes to model the error
structure of earnings in a longitudinal data analysis", Journal of Econometrics,
18, pp82-114
[17] Mirrlees, J. A., 1971, "Optimal tax theory: a synthesis", Review of Economic
Studies, pp. 327-358.
[18] Scholz, J. K., 1996, "In work bene�ts in the United States: the earned income
tax credit", Economic Journal, 106(434), pp. 156-169
[19] Townsend, R.M., 1994, "Risk and insurance in village India", Econometrica, 62,
no.3, pp. 539-591.
[20] Varian, H., 1980, "Redistributive taxation as social insurance", Journal of Public
Economics, 14, pp. 49-68
30
Table 1: Thresholds for current federal tax brackets
Tax Rate Tax Bracket
(%) single married jointly married seperately
15 0 0 0
28 26,250 43,850 21,925
31 63,550 105,950 52,975
36 132,660 161,450 80,725
39.6 288,350 288,350 144,175
31
Table 2: Proportion paying at each marginal federal tax rate for 1982-1998
Year Tax Bracket % Paying Rate Year Tax Bracket % Paying Rate
1982 0 13.410-20 19.521-30 30.831-40 25.241-50 11.1
1983 0 14.6 1984 0 15.810-20 26.8 10-20 28.221-30 31.6 21-30 29.031-40 21.9 31-40 19.741-50 5.2 41-50 7.4
1985 0 15.2 1986 0 15.510-20 26.3 10-20 26.621-30 28.4 21-30 28.431-40 22.5 31-40 22.341-50 7.7 41-50 7.3
1987 0 17.1 1988 0 17.011 2.915 38.0 15 42.428 23.3 28 40.535 15.938.5 2.6
1989 0 17.2 1990 0 17.415 42.4 15 42.328 40.2 28 40.2
1991 0 17.1 1992 0 18.615 43.9 15 43.028 33.2 28 31.531 5.6 31 6.7
1993 0 18.0 1994 0 17.815 44.9 15 43.728 30.8 28 32.631 5.5 31 5.136 0.5 36 0.6
1995 0 17.6 1996 0 19.415 43.2 15 42.128 33.3 28 32.631 5.1 31 4.336 0.5 36 1.2
39.6 0.1
1997 15 61.1 1998 15 58.228 32.1 28 34.231 4.9 31 5.236 1.4 36 1.839.6 0.2 39.6 0.3
32
Table 3: State Individual Income Tax Rates in the US
State Tax Rates Exemptions
low high single married dependents
Alabama 2.0 5.0 1,500 3,000 300
Alaska no state tax
Arizona 2.87 5.04 2,100 4,200 2,300
Arkansas 1.0 7.0 20* 40* 20*
California 1.0 9.3 72* 142* 227*
Colorado 4.63 4.63 none
Connecticut 3.0 4.5 12,000 24,000 0
Delaware 2.2 5.95 110* 220* 110*
Florida no state tax
Georgia 1.0 6.0 2,700 5,400 2,700
Hawaii 1.5 8.5 1,040 2,080 1,040
Idaho 2.0 8.2 2,900 5,800 2,900
Illinois 3.0 3.0 2,000 4,000 2,000
Indiana 3.4 3.4 1,000 2,000 1,000
Iowa 0.36 8.98 40* 80* 40*
Kansas 3.5 6.45 2,250 4,500 2,250
Kentucky 2.0 6.0 20* 40* 20*
Louisiana 2.0 6.0 4,500 9,000 1,000
Maine 2.0 8.5 2,850 5,700 2,850
Maryland 2.0 4.75 1,850 3,700 1,850
Massachusetts 5.6 5.6 4,400 8,800 1,000
Michigan 4.2 4.2 2,800 5,600 2,800
Minnesota 5.35 7.85 2,900 5,800 2,900
Mississippi 3.0 5.0 6,000 12,000 1,000
Missouri 1.5 6.0 2,100 4,200 2,100
Montana 2.0 11.0 1,610 3,220 1,610
*Tax Credits.
33
Table 3: (cont.) State Individual Income Tax Rates in the US
State Tax Rates Exemptions
low high single married dependents
Nebraska 2.51 6.68 91* 182* 91*
Nevada no state tax
New Hampshire taxes unearned income only
New Jersey 1.4 6.37 1,000 2,000 1,500
New Mexico 1.7 8.2 2,900 5,800 2,900
New York 4.0 6.85 - - 1,000
North Carolina 6.0 7.75 2,500 5,000 2,500
North Dakota 2.67 12.0 2,900 5,800 2,900
Ohio 0.691 6.98 1,050 2,100 1,050
Oklahoma 0.5 6.75 1,000 2,000 1,000
Oregon 5.0 9.0 132* 264* 132*
Pennsylvania 2.8 2.8 none
Rhode Island 25.5% of federal taxes
South Carolina 2.5 7.0 2,900 5,800 2,900
South Dakota no state tax
Tennessee taxes unearned income only
Texas no state tax
Utah 2.3 7.0 2,175 4,350 2,174
Vermont 24% of federal taxes
Virginia 2.0 5.75 800 1,600 800
Washington no state tax
West Virginia 3.0 6.5 2,000 4,000 2,000
Wisconsin 4.6 6.75 700 1,400 400
Wyoming no state tax
Dist. Columbia 5.0 9.0 1,370 2,740 1,370
*Tax Credits.
34
Table 4: The level of transfers in the US
transfer average average if received % receive
social security 247 6,710 3.6
supplementary security income 74 3,328 2.2
unemployment compensation 160 2,439 6.5
worker's compensation 117 4,484 2.6
welfare 169 3,768 4.4
pension 266 8,815 3.0
food stamps 101 1,918 5.3
total � 871 4,680 18.6
�Excluding pension income
Table 5: Measuring tax progressivity
mean mean
marginal rate tax bracket �
Federal 19.2 20.2
State:
Overall 3.7 4.2 32.0
California 5.0 5.3 36.4
Florida - - 27.6
New York 6.3 7.4 36.8
Ohio 3.8 4.0 33.0
Pennsylvania 2.2 2.4 29.8
Texas - - 28.9
35
Table 6: Regressing �ksdj (cit) against the mean tax rate (standard errors in paren-
thesis).
basic control Instrument
�t�1 �t�1 and �t�2 �t�2
Marginal Tax Rate
� 1.633��� 1.636��� 1.590�� 2.146�� 2.126��
(0.590) (0.583) (0.775) (0.854) (0.846)
constant 0.177 0.174 0.189 0.081 0.086
(0.127) (0.125) (0.172) (0.184) (0.182)
Tax Bracket
� 1.036�� 1.054�� 0.805 1.163 1.294�
(0.521) (0.515) (0.668) (0.754) (0.708)
constant 0.294�� 0.288�� 0.349�� 0.277 0.247
(0.126) (0.124) (0.163) (0.179) (0.167)
N 185 185 176 166 166
Table 7: Regressing �sdj (ln cit) against the mean tax rate (standard errors in paren-
thesis).
basic control Instrument
�t�1 �t�1 and �t�2 �t�2
Marginal Tax Rate
� 0.396� 0.402�� 0.481�� 0.574�� 0.678���
(0.198) (0.194) (0.207) (0.232) (0.233)
constant 0.390��� 0.379��� 0.358��� 0.339��� 0.317���
(0.042) (0.041) (0.045) (0.049) (0.050)
Tax Bracket
� 0.391�� 0.428�� 0.377�� 0.423�� 0.512���
(0.172) (0.168) (0.148) (0.172) (0.175)
constant 0.379��� 0.367��� 0.376��� 0.366��� 0.346���
(0.039) (0.038) (0.035) (0.039) (0.040)
N 185 185 176 166 166
36
Table 8: Regressing �ksdj (cit) against the �j (standard errors in parenthesis).
basic control Instrument
�t�1 �t�1 and �t�2 �t�2
Taxes from CEX
� 1.309 1.322 1.575 2.422 2.781
(0.845) (0.843) (1.494) (1.715) (1.928)
constant 0.229 0.226 0.169 -0.021 -0.106
(0.203) (0.202) (0.359) (0.409) (0.460)
Taxes only
� 1.910�� 1.912�� 1.851 2.812�� 2.995��
(0.829) (0.826) (1.382) (1.368) (1.414)
constant 0.098 0.097 0.117 -0.091 -0.133
(0.190) (0.189) (0.323) (0.314) (0.322)
Taxes and Transfers
� 1.459�� 1.459�� 1.452 1.838 1.629
(0.724) (0.721) (1.368) (1.345) (1.351)
constant 0.166 0.166 0.172 0.084 0.137
(0.185) (0.184) (0.356) (0.347) (0.349)
N 188 188 179 169 169
37
Table 9: Regressing �ksdj (ln cit) against the �j (standard errors in parenthesis).
basic control Instrument
�t�1 �t�1 and �t�2 �t�2
Taxes from CEX
� 0.374�� 0.395�� 0.619�� 0.627�� 1.161��
(0.181) (0.182) (0.279) (0.279) (0.485)
constant 0.381��� 0.376��� 0.319��� 0.317��� 0.192�
(0.042) (0.042) (0.067) (0.067) (0.115)
Taxes only
� 0.520��� 0.519��� 0.589�� 0.624�� 0.633�
(0.184) (0.184) (0.279) (0.287) (0.356)
constant 0.350��� 0.350��� 0.331��� 0.324��� 0.322���
(0.041) (0.041) (0.065) (0.065) (0.081)
Taxes and Transfers
� 0.493��� 0.490��� 0.634��� 0.653��� 0.611��
(0.185) (0.185) (0.243) (0.251) (0.305)
constant 0.344��� 0.344��� 0.304��� 0.300��� 0.311���
(0.046) (0.046) (0.062) (0.063) (0.077)
N 186 186 176 166 166
38
Table 10: Estimated parameters in the insurance function � (�) = �0 + �1� + error
from tables 7 and 9.
basic control Instrument
�t�1 �t�1 and �t�2 �t�2
Marginal tax rate
�0 0.33 0.35 0.38 0.41 0.45
�1 -0.67 -0.69 -0.82 -0.98 -1.16
Tax Bracket
�0 0.35 0.37 0.35 0.37 0.40
�1 -0.67 -0.73 -0.64 -0.72 -0.87
Incomes from the CEX
�0 0.34 0.35 0.45 0.45 0.67
�1 -0.64 -0.67 -1.06 -1.07 -1.99
Taxes only using TAXSIM
�0 0.39 0.39 0.43 0.44 0.44
�1 -0.89 -0.89 -1.01 -1.07 -1.08
Taxes and Transfers using TAXSIM
�0 0.41 0.41 0.47 0.48 0.46
�1 -0.84 -0.84 -1.08 -1.11 -1.04
39